Hydrologic evaluation of a Generalized Statistical Uncertainty Model for Satellite Precipitation Products

Friday, 19 December 2014: 11:50 AM
Sepideh Sarachi, Kuo-lin Hsu and Soroosh Sorooshian, University of California Irvine, Irvine, CA, United States
Development of satellite based precipitation retrieval algorithms and using them in hydroclimatic studies have been of great interest to hydrologists. It is important to understand the uncertainty associated with precipitation products and how they further contribute to the variability in stream flow simulation. In this study a mixture model of Generalized Normal Distribution and Gamma distribution (GND-G) is used to model the joint probability distribution of satellite-based (PERSIANN) and stage IV radar rainfall. The study area for constructing the uncertainty model covers a 15°×15°box of 0.25°×0.25° cells over the eastern United States for summer 2004 to 2009. Cells are aggregated in space and time to obtain data with different resolutions for the construction of the model’s parameter space.

This uncertainty model is evaluated using data from National Weather Service (NWS) Distributed Hydrologic Model Intercomparison Project - Phase 2 (DMIP 2) basin over Illinois River basin south of Siloam, OK. This data covers the time period of 2006 to 2008.The uncertainty range of precipitation is estimated. The impact of precipitation uncertainty to the stream flow estimation is demonstrated by Monte Carlo simulation of precipitation forcing in the Sacramento Soil Moisture Accounting (SAC-SMA) model. The results show that using precipitation along with its uncertainty distribution as forcing to SAC-SMA make it possible to have an estimation of the uncertainty associated with the stream flow simulation ( in this case study %90 confidence interval is used). The mean of this stream flow confidence interval is compared to the reference stream flow for evaluation of the model and the results show that this method helps to better estimate the variability of the stream flow simulation along with its statistics e.g. percent bias and root mean squared error.